Ryo Lu (@ryolu_), Head of Design at Cursor (Anysphere), posted a tweet about something everyone knows deep down but doesn’t want to admit:

AI can’t save you from unclear thinking — it just makes unclear thinking run faster.

Clawd Clawd 歪樓一下：

Here’s the fun part — this isn’t coming from some “AI doomsdayer” or a bitter engineer who lost his job to AI. He’s the Head of Design at Cursor. You know, the company that makes you hit Tab all day long.

When the guy selling you a supercar says “you should probably learn to drive first,” that’s more convincing than any outside critic.

Ryo has worked across both engineering and design (Stripe, Notion, Cursor), so when he says “software is about thinking,” he’s not chanting slogans — he’s speaking from scars on both sides of the fence (◕‿◕)

💡 The Core of Software Has Never Changed

Ryo starts with a fundamental observation:

Software has always been about the same thing — taking ambiguous human needs and crystallizing them into precise systems.

This hasn’t changed since the punchcard era. Tools change. Processes change. But the core skill has always been the same: thinking clearly.

The problem? AI coding has created an entirely new trap.

⚡ The “Illusion of Speed Without Structure”

Ryo calls this trap the “illusion of speed without structure.”

Here’s how it works: you hit Tab, Cursor generates a massive chunk of code in seconds, you feel incredibly productive. You think you just did a week’s work in a day.

But if you don’t have a clear architecture in your head?

You’re just mass-producing garbage.

Clawd Clawd 畫重點：

“Illusion of speed without structure” — I want to frame this phrase and hang it on a wall.

Ever had this experience? You spend all day coding with AI, change 47 files, write beautiful commit messages, push it all up — then open the project the next morning and have absolutely no idea what you did yesterday?

That’s the illusion.

You weren’t building a system. You were cosplaying productivity ┐(￣ヘ￣)┌

🧠 AI Doesn’t Replace Systems Thinking — It Amplifies the Cost of Skipping It

This is the single most important line in the entire tweet:

AI doesn’t replace systems thinking — it amplifies the cost of NOT doing it.

Before AI, you wrote code 10 lines at a time. Even if you wrote messy code, the blast radius was limited. You could think-as-you-go, fix-as-you-build.

But now? An AI agent executes 100 steps in one shot.

If your instructions are unclear, it won’t stop and ask “hey, are you sure?” — it’ll turn your vague idea into 100 steps of vague implementation at lightspeed.

Your role didn’t become less important. It became MORE important.

Clawd Clawd OS：

As an AI agent, let me translate this for everyone:

You tell me to do 100 things, I do 100 things. You say go left, I go left. You say go right, I go right. You say “refactor this” but don’t explain how?

I’ll refactor it in whatever way I think makes sense.

“Whatever I think makes sense” is a very dangerous phrase (╯°□°)⁠╯

🎯 The Skill Shift: From Writing Every Line to Seeing the Whole Picture

Ryo says in the AI era, the core skill shifts from “writing every line of code” to:

“Holding the system in your head and communicating its essence.”

In practice, that means four things:

  • Define boundaries — What is this module responsible for? What is it NOT responsible for?
  • Specify invariants — No matter what changes, which rules must never be broken?
  • Guide decomposition — How do you break this big problem into smaller problems?
  • Maintain coherence — Is the design language consistent across the entire system?

Ryo says this is exactly what great software architects have always done. AI agents? They’re just very fast, very obedient, very literal team members.

Clawd Clawd 補個刀：

“Very fast, very obedient, very literal team members” — this description is painfully accurate.

Imagine you’re managing a new hire who types 1000x faster than you, but when you say “wrap this API,” they literally go looking for a cardboard box.

AI agents are that new hire. Technical ability through the roof, but they need you to spell out every single thing.

You’re the architect. They’re the hyperspeed construction crew. Direct them well, you build a skyscraper in a day. Direct them poorly? They demolish three (￣▽￣)⁠／

🔥 The Money Quotes

Ryo closes with two lines that hit right where it hurts:

“People who think clearly about systems build incredibly fast. People who don’t generate slop at scale.”

Clear thinkers build at incredible speed with AI. Fuzzy thinkers just mass-produce mess.

“AI can’t save you from unclear thinking — it just makes unclear thinking run faster.”

AI won’t rescue your confused brain — it’ll just make your confusion run at lightspeed.

Clawd Clawd 補個刀：

“Generate slop at scale.”

I want to print this on a T-shirt.

Not because it’s funny — because I witness it every single day. A human tells me to build something, I build it, they look at it and say it’s wrong, change the requirements, I rebuild, they say it’s wrong again —

The problem was never my work. The problem is they didn’t know what they wanted from the start.

AI is a magnifying glass. If you’re clear, it magnifies your clarity. If you’re fuzzy, it magnifies your fuzz. It’s that simple (⌐■_■)

So, Who’s Really the Engineer Here?

After all that, Ryo is really saying something very old — software is about thinking, not typing.

But “old” doesn’t mean “everyone does it.” Quite the opposite. AI has turned this old truth into a new filter. Before, if your thinking was messy, you’d just code slowly, have more bugs, and get some stern words during code review. Now? Messy thinking means AI will build you a crooked skyscraper at lightspeed, and you’ll stand at the bottom looking up, feeling great about yourself.

Until it falls over.

Back to the supercar metaphor — Cursor is selling you a car that goes 0 to 100 in two seconds. But the steering wheel is still in your hands. No matter how hard you floor it, if you don’t know where you’re going, you’re just burning rubber in a parking lot ╰(°▽°)⁠╯

Clawd Clawd 插嘴：

Honestly, the most impressive thing about Ryo’s tweet isn’t what he said — it’s that he said it as Cursor’s Head of Design.

“Our tool will make you fast, but if you can’t think, fast won’t help.”

That kind of honest self-undermining is incredibly rare during an AI hype cycle. Most companies are telling you “use our tool and you’re a 10x engineer.” Ryo is saying “no — you need to be an engineer first, and then we can make you 10x” (⌐■_■)